#IMPORT (wrangled) data 
df_raw <- read_csv("data/CLEAN_coded_utterances.csv") 
df_reps <- read_csv("data/CLEAN_coded_utterances.csv") 

#WRANGLE into DF of coded utterances 
#NOT unique utterances, 1 obs for each utterance+detail-code
df_coded <- df_raw %>% 
  #rename and factorize cols
  mutate(
    #UNIQUE IDS
    sid = factor(SID), #unique ID for utterance+detail-code
    pid = factor(PID, levels = c( #define level order so happiness first
      #HAPPINESS-FIRST    
      "bjs827ee1u", "3r2sh20ei", "4728sjuiz","7ACC0B75","92ghd48xe","iurmer289", "s294hoei",
      #SPACE-FIRST    
      "j2719eertu2","lkin27js09b","li832lin23","7382kwtue","E1D39056","8v892iige")),   
    #create unique ID for utterances
    uid = factor(as.numeric(factor(paste(pid,factor(Utterance))))), #construct a unique ID for utterances
    #recode lower case and order based on true task order
    TASK = factor(recode(Condition, "Static"="static", "Interactive"="ixn" )),
    TASK = factor(TASK, levels = c("static", "ixn")), #reorder factor levels
    #rename Notebook as DATASET
    DATASET = factor(recode(Notebook, "Happiness"="happiness", "Space"="space")),
    #create temp dataset order var
    data_order = factor(paste(TASK,"_",DATASET)), #create an order var 
    data_order = recode(data_order, "ixn _ happiness"="space-first",
                                    "ixn _ space"="happiness-first",
                                    "static _ happiness"="happiness-first",
                                    "static _ space"="space-first"),
    utterance = Utterance,
    reps_group = factor(Final_Group),
    reps_all = factor(`All representations`),
    #rename flags
    flag_story = `Dylan Flag Storytelling`,
    flag_correction = `Dylan Flag Correction`,
    flag_simultaneous = `Dylan Flag Simultaneous Characterization`,
    #recode and order TOP LEVEL CODES 
    code_topic = factor(Highlevel),
    code_topic = recode(code_topic, "ANALYSIS PROCESS" = "PROCESS"),
    code_topic = factor(code_topic, levels = c("PROCESS","DATASET","VARIABLE","RELATIONSHIP")),
    code_datatype = factor(`Data Type`),
    code_detail = factor(`Utterance Type`),
    timestamp = adj_timestamp,
    ixn = factor(interaction_used), #was interaction used?
    PNUM = factor(PNUM,levels = c("P6", "P9", "P10", "P2", "P4", "P12","P13", 
                                   "P5", "P7", "P8", "P3", "P1","P11")),
    
    ) %>% 
  dplyr::select(sid,pid,PNUM,uid,TASK,DATASET,timestamp,ixn,code_topic,code_detail,code_datatype,
         flag_story, flag_correction, flag_simultaneous, utterance, reps_group, reps_all, data_order) %>% 
  arrange(data_order)

#REPLACE NA in logicals to FALSE  
df_coded$flag_story[is.na(df_coded$flag_story)] <- FALSE
df_coded$flag_correction[is.na(df_coded$flag_correction)] <- FALSE
df_coded$flag_simultaneous[is.na(df_coded$flag_simultaneous)] <- FALSE



#CALCULATE RELATIVE TASK TIMES
df_time <- df_coded %>% mutate(
  time = hms::as_hms(timestamp)
) %>% group_by(pid, TASK) %>% 
  # dplyr::summarise( .groups="keep",
  mutate(
    task_start = hms::as_hms(min(time)),
    task_end = hms::as_hms(max(time)),
    task_mins = round(difftime(task_end,task_start, units="mins"),1),
    task_second = task_end - task_start,
    relative_time_s = timestamp-task_start,
    relative_time = as.double(relative_time_s)
  ) %>% ungroup() %>% 
  dplyr::select(pid,PNUM, code_topic,code_detail, TASK,DATASET,timestamp,task_start,relative_time_s,relative_time)

There are 742 rows in the df_coded dataset, where each row represents an utterance coding (i.e. utterance + detail code). There are 662 unique utterances. The difference indicates utterances that were dual-coded (i.e. two detail-level codes). No more than two codes were applied to a single utterance. For the purposes of analysis, dual-coded utterances will be treated as two utterances, as they have two distinct (but lexically insepeperable) units of meaning.

1 DATA PROFILE

df_coded%>% summarytools::dfSummary(
             plain.ascii  = FALSE,
             graph.magnif = 0.75,
             style        = "grid",
             tmp.img.dir  = "temp",
             missing.col = FALSE, 
             method = "render"
)

1.0.1 Data Frame Summary

1.0.1.1 df_coded

Dimensions: 742 x 18
Duplicates: 0

No	Variable	Stats / Values	Freqs (% of Valid)	Valid	Missing
1	sid [factor]	1. 0 2. 1 3. 2 4. 3 5. 4 6. 5 7. 6 8. 7 9. 8 10. 9 [ 732 others ]	1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 732 (98.7%)	742 (100.0%)	0 (0.0%)
2	pid [factor]	1. bjs827ee1u 2. 3r2sh20ei 3. 4728sjuiz 4. 7ACC0B75 5. 92ghd48xe 6. iurmer289 7. s294hoei 8. j2719eertu2 9. lkin27js09b 10. li832lin23 [ 3 others ]	29 ( 3.9%) 103 (13.9%) 43 ( 5.8%) 28 ( 3.8%) 56 ( 7.5%) 87 (11.7%) 88 (11.9%) 82 (11.1%) 48 ( 6.5%) 51 ( 6.9%) 127 (17.1%)	742 (100.0%)	0 (0.0%)
3	PNUM [factor]	1. P6 2. P9 3. P10 4. P2 5. P4 6. P12 7. P13 8. P5 9. P7 10. P8 [ 3 others ]	29 ( 3.9%) 103 (13.9%) 43 ( 5.8%) 28 ( 3.8%) 56 ( 7.5%) 87 (11.7%) 88 (11.9%) 82 (11.1%) 48 ( 6.5%) 51 ( 6.9%) 127 (17.1%)	742 (100.0%)	0 (0.0%)
4	uid [factor]	1. 1 2. 2 3. 3 4. 4 5. 5 6. 6 7. 7 8. 8 9. 9 10. 10 [ 652 others ]	2 ( 0.3%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 2 ( 0.3%) 1 ( 0.1%) 1 ( 0.1%) 730 (98.4%)	742 (100.0%)	0 (0.0%)
5	TASK [factor]	1. static 2. ixn	403 (54.3%) 339 (45.7%)	742 (100.0%)	0 (0.0%)
6	DATASET [factor]	1. happiness 2. space	431 (58.1%) 311 (41.9%)	742 (100.0%)	0 (0.0%)
7	timestamp [hms, difftime]	min : 622 med : 2857 max : 6900 units : secs	622 distinct values	742 (100.0%)	0 (0.0%)
8	ixn [factor]	1. FALSE 2. TRUE	633 (85.3%) 109 (14.7%)	742 (100.0%)	0 (0.0%)
9	code_topic [factor]	1. PROCESS 2. DATASET 3. VARIABLE 4. RELATIONSHIP	160 (21.6%) 176 (23.7%) 122 (16.4%) 284 (38.3%)	742 (100.0%)	0 (0.0%)
10	code_detail [factor]	1. data orientation 2. data provenance 3. data size 4. distribution outlier (var 5. distribution range [min, 6. distribution shape [shape 7. distribution variance (sd 8. missing data 9. outlier (relationship) 10. plan of action [ 8 others ]	16 ( 2.2%) 11 ( 1.5%) 9 ( 1.2%) 9 ( 1.2%) 33 ( 4.4%) 79 (10.6%) 1 ( 0.1%) 76 (10.2%) 20 ( 2.7%) 52 ( 7.0%) 436 (58.8%)	742 (100.0%)	0 (0.0%)
11	code_datatype [factor]	1. distribution (continuous 2. distribution (categorical 3. relationship (categorical 4. relationship (categorical 5. relationship (continuous 6. relationship (multivariat	76 (17.8%) 54 (12.7%) 28 ( 6.6%) 55 (12.9%) 146 (34.3%) 67 (15.7%)	426 (57.4%)	316 (42.6%)
12	flag_story [logical]	1. FALSE 2. TRUE	700 (94.3%) 42 ( 5.7%)	742 (100.0%)	0 (0.0%)
13	flag_correction [logical]	1. FALSE 2. TRUE	733 (98.8%) 9 ( 1.2%)	742 (100.0%)	0 (0.0%)
14	flag_simultaneous [logical]	1. FALSE 2. TRUE	682 (91.9%) 60 ( 8.1%)	742 (100.0%)	0 (0.0%)
15	utterance [character]	1. [Talking about the profil 2. actually, let me see if p 3. Although we have like les 4. And are they within range 5. And confidence in governm 6. And just I want to see ho 7. And so it looks like it s 8. And then if I had more ti 9. Because it does seem like 10. Data frame. Got a bunch o [ 652 others ]	2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 722 (97.3%)	742 (100.0%)	0 (0.0%)
16	reps_group [factor]	1. barplot 2. columns 3. columns_data_dictionary 4. data_dictionary 5. data_dictionary_dataframe 6. data_dictionary_describe 7. dataframe 8. dataframe_describe 9. dataframe_heatmap 10. dataframe_pairplot [ 15 others ]	16 ( 2.2%) 4 ( 0.5%) 1 ( 0.1%) 56 ( 7.5%) 1 ( 0.1%) 9 ( 1.2%) 76 (10.2%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 576 (77.6%)	742 (100.0%)	0 (0.0%)
17	reps_all [factor]	1. affect_corruption_brush_7 2. Age_CryoSleep_scatterplot 3. age_CryoSleep_ShoppingMal 4. Age_RoomService_scatterpl 5. age_roomservice_scatterpl 6. Age_RoomService_scatterpl 7. Age_ShoppingMall_scatterp 8. altair_profile_contVars_j 9. alx_barplot_df_homeplanet 10. alx_barplot_df_homeplanet [ 245 others ]	4 ( 0.6%) 3 ( 0.4%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 2 ( 0.3%) 1 ( 0.1%) 693 (97.7%)	709 (95.6%)	33 (4.4%)
18	data_order [factor]	1. space-first 2. happiness-first	308 (41.5%) 434 (58.5%)	742 (100.0%)	0 (0.0%)

ARF has reviewed data profile for missing data and correct factorization.

2 UTTERANCES

Utterances are the lowest-level discrete units of meaning transcribed from the EDA Task transcripts. Utterances are coded at two levels of analysis: (1) topic-code gives a high level topic of the participant’s verbalization, (2) detail-code gives the lower level detail of the subject.

In the following subsections we explore the distribution of number of utterances based on TASK, DATASET, and PARTICIPANT, before describing the distribution of utterances through the timecourse of the TASK.

2.1 [Aggregated] Utterances

FIRST we explore the distribution of utterances by Analysis Task, Dataset, Participant and Time, irrespective of what the utterance was about (topic, detail).

RQ: How much did participants talk aloud during EDA? When did they talk aloud?

Answer: Inspection of frequency tables and visualizations suggests that the most substantial determinant of how many utterances an individual made is individual participant-level differences, rather than structural differences imposed by the TASK or DATASET. This is not altogether unexpected given the fact that across both tasks (static/interactive) and datasets the structure of the experimental task was the same

2.1.1 by TASK

print("BY TASK")

[1] “BY TASK”

freq(df_coded$TASK, 
     cumul      = FALSE,
     headings   = FALSE,
     report.nas = FALSE,
     plain.ascii = FALSE)

	Freq	%
static	403	54.31
ixn	339	45.69
Total	742	100.00

2.1.2 by TASK and DATASET

#COUNT BY TASK AND DATASET
ctable(x = df_coded$TASK, 
       y = df_coded$DATASET, 
       prop = "t")

Cross-Tabulation, Total Proportions
TASK * DATASET
Data Frame: df_coded

	DATASET	happiness	space	Total
TASK
static		263 (35.4%)	140 (18.9%)	403 ( 54.3%)
ixn		168 (22.6%)	171 (23.0%)	339 ( 45.7%)
Total		431 (58.1%)	311 (41.9%)	742 (100.0%)

#DF SUMMARIZED BY TASK + DATASET
df_summary <- df_coded %>% 
  group_by(TASK,DATASET) %>% 
  dplyr::summarise(
    c = n()
  )

#STACKED BAR BY TASK
ggplot(df_summary, aes(x = TASK, y=c, fill= DATASET)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  # scale_fill_brewer(type="qual", palette = 4) +
  labs( title = "Utterances by TASK and DATASET",
        subtitle = "More utterances in STATIC; more utterances in HAPPINESS",
        x= "TASK", y = "count") + theme_minimal()

# + theme(legend.position = "blank")

2.1.3 by PARTICIPANT

#COUNT BY PARTICIPANT AND TASK
ctable(x = df_coded$PNUM, 
       y = df_coded$TASK, 
       prop = "r")

Cross-Tabulation, Row Proportions
PNUM * TASK
Data Frame: df_coded

	TASK	static	ixn	Total
PNUM
P6		11 (37.9%)	18 (62.1%)	29 (100.0%)
P9		63 (61.2%)	40 (38.8%)	103 (100.0%)
P10		30 (69.8%)	13 (30.2%)	43 (100.0%)
P2		18 (64.3%)	10 (35.7%)	28 (100.0%)
P4		35 (62.5%)	21 (37.5%)	56 (100.0%)
P12		46 (52.9%)	41 (47.1%)	87 (100.0%)
P13		60 (68.2%)	28 (31.8%)	88 (100.0%)
P5		33 (40.2%)	49 (59.8%)	82 (100.0%)
P7		29 (60.4%)	19 (39.6%)	48 (100.0%)
P8		17 (33.3%)	34 (66.7%)	51 (100.0%)
P3		24 (44.4%)	30 (55.6%)	54 (100.0%)
P1		10 (40.0%)	15 (60.0%)	25 (100.0%)
P11		27 (56.2%)	21 (43.8%)	48 (100.0%)
Total		403 (54.3%)	339 (45.7%)	742 (100.0%)

#UTTERANCES by PARTICPANT facet TASK
gf_bar( PNUM ~., fill = ~ DATASET, data = df_coded) %>% 
  gf_facet_grid(.~TASK) + 
  labs(
    title = "Utterances by Participant, Dataset and Task",
    subtitle = "",
    x = "number of coded utterances",
    y = "participant",
    fill = "DATASET"
  )

2.1.4 through TIME

#DOTPLOT
ggplot(df_time, aes(x=relative_time, y = PNUM)) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time$TASK) +
  scale_color_brewer(type="qual", palette = 3) +
  theme_minimal() + labs(
    title = "Participant Utterances over timecourse of Task",
    x= "timecourse of task (seconds)", y = "Participant",
    color = "Topic"
  )

#HISTOGRAMS BY TASK
ggplot(df_time, aes(x = relative_time)) + 
  geom_histogram(binwidth = 30,aes(y=..density..)) + 
  geom_density()+
  facet_grid(df_time$TASK) +
  theme_minimal() + labs(
    title = "Participant Utterances over timecourse of Task",
    x= "timecourse of task (seconds)", y = "frequency of utterances",
  ) + theme_minimal() + theme(legend.position = "blank")

## Warning: The dot-dot notation (`..density..`) was deprecated in ggplot2 3.4.0.
## ℹ Please use `after_stat(density)` instead.

2.2 [TOPIC of] Utterances

NEXT we explore the distribution of utterances coded by high level TOPIC, across Analysis Task, Dataset, Participant and Time.

RQ: What kinds of things did participants talk aloud during EDA? Did they progress through any ‘topical phases’ over the course of the task? Or are topics equally distributed across analysis time?

Answer: Inspection of frequency tables and visualizations suggests that:

Individual differences continue to play an important role
There do not appear to be strong TASK/DATASET effects on topic that are consistent across participants.
PROCESS and RELATIONSHIP topics are more evenly distributed across the timecourse of analysis, while DATASET AND VARIABLE topics are more tightly clustered near the beginning of the analysis. This pattern of distribution is sensical given what we know about EDA, and is consistent with the intuition that patterns of thought during EDA are likely more iterative and situational than we think (or model).

2.2.1 by TASK

#COUNT BY TASK
ctable(x = df_coded$code_topic, 
       y = df_coded$TASK, 
       prop = "r")

Cross-Tabulation, Row Proportions
code_topic * TASK
Data Frame: df_coded

	TASK	static	ixn	Total
code_topic
PROCESS		92 (57.5%)	68 (42.5%)	160 (100.0%)
DATASET		100 (56.8%)	76 (43.2%)	176 (100.0%)
VARIABLE		77 (63.1%)	45 (36.9%)	122 (100.0%)
RELATIONSHIP		134 (47.2%)	150 (52.8%)	284 (100.0%)
Total		403 (54.3%)	339 (45.7%)	742 (100.0%)

#DF SUMMARIZED BY TASK + DATASET
df_summary <- df_coded %>% 
  group_by(code_topic, TASK) %>% 
  dplyr::summarise(
    c = n()
  )

#STACKED BAR BY TASK
ggplot(df_summary, aes(x = TASK, y=c, fill= fct_rev(code_topic))) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="qual", palette = 3) +
  labs( title = "TOPICS by TASK",
        subtitle = "",
        x= "TASK", y = "count", fill="TOPIC") + theme_minimal()

# + theme(legend.position = "blank")

2.2.2 by TASK and DATASET

#DF SUMMARIZED BY TASK + DATASET
df_summary <- df_coded %>%
  group_by(code_topic, TASK,DATASET) %>%
  dplyr::summarise(
    c = n()
  )

#STACKED BAR BY TASK FACET DATASET
ggplot(df_summary, aes(x = TASK, y=c, fill= fct_rev(code_topic))) +
  facet_wrap(df_summary$DATASET) +
  geom_col() +
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") +
  scale_fill_brewer(type="qual", palette = 3) +
  labs( title = "TOPICS by TASK and DATASET",
        subtitle = "",
        x= "TASK", y = "count", fill="TOPIC") + theme_minimal()

# + theme(legend.position = "blank")

2.2.3 by PARTICIPANT

#COUNT BY PARTICIPANT 
ctable(x = df_coded$PNUM, 
       y = df_coded$code_topic, 
       prop = "r")

Cross-Tabulation, Row Proportions
PNUM * code_topic
Data Frame: df_coded

	code_topic	PROCESS	DATASET	VARIABLE	RELATIONSHIP	Total
PNUM
P6		6 (20.7%)	5 (17.2%)	5 (17.2%)	13 (44.8%)	29 (100.0%)
P9		19 (18.4%)	36 (35.0%)	23 (22.3%)	25 (24.3%)	103 (100.0%)
P10		8 (18.6%)	11 (25.6%)	8 (18.6%)	16 (37.2%)	43 (100.0%)
P2		3 (10.7%)	8 (28.6%)	16 (57.1%)	1 ( 3.6%)	28 (100.0%)
P4		10 (17.9%)	6 (10.7%)	11 (19.6%)	29 (51.8%)	56 (100.0%)
P12		21 (24.1%)	28 (32.2%)	17 (19.5%)	21 (24.1%)	87 (100.0%)
P13		41 (46.6%)	7 ( 8.0%)	8 ( 9.1%)	32 (36.4%)	88 (100.0%)
P5		6 ( 7.3%)	18 (22.0%)	18 (22.0%)	40 (48.8%)	82 (100.0%)
P7		14 (29.2%)	14 (29.2%)	7 (14.6%)	13 (27.1%)	48 (100.0%)
P8		10 (19.6%)	7 (13.7%)	3 ( 5.9%)	31 (60.8%)	51 (100.0%)
P3		7 (13.0%)	6 (11.1%)	2 ( 3.7%)	39 (72.2%)	54 (100.0%)
P1		8 (32.0%)	4 (16.0%)	0 ( 0.0%)	13 (52.0%)	25 (100.0%)
P11		7 (14.6%)	26 (54.2%)	4 ( 8.3%)	11 (22.9%)	48 (100.0%)
Total		160 (21.6%)	176 (23.7%)	122 (16.4%)	284 (38.3%)	742 (100.0%)

#TOPICS by PARTICPANT facet TASK
(p <- gf_bar( PNUM ~., fill = ~ fct_rev(code_topic), data = df_coded) %>% 
  gf_facet_grid(.~TASK) + 
  scale_fill_brewer(type="qual", palette = 3) +
  labs(
    title = "Utterances by Participant, Dataset and Task",
    subtitle = "",
    x = "number of coded utterances",
    y = "participant",
    fill = "TOPIC"
  ) + theme_minimal())

ggsave(p, file="figures/topics_by_count.png")



# #TOPICS by PARTICPANT facet TASK
# gf_bar( PNUM ~., fill = ~ fct_rev(code_topic), data = df_coded) %>% 
#   gf_facet_grid(DATASET~TASK) + 
#   scale_fill_brewer(type="qual", palette = 3) +
#   labs(
#     title = "Utterances by Participant, Dataset and Task",
#     subtitle = "",
#     x = "number of coded utterances",
#     y = "participant",
#     fill = "DATASET"
#   )

2.2.4 by TIME

#HISTOGRAMS BY TASK
ggplot(df_time, aes(x = relative_time)) + 
  geom_histogram(binwidth = 30,aes(y=..density.., fill = fct_rev(code_topic), color = fct_rev(code_topic))) + 
  geom_density()+
  facet_grid(df_time$code_topic ~ df_time$TASK) +
  scale_fill_brewer(type="qual", palette = 3) +
  scale_color_brewer(type="qual", palette = 3) +
  theme_minimal() + labs(
    title = "Topic of Utterance over timecourse of Task",
    x= "timecourse of task (seconds)", y = "frequency of utterances",
    fill = "Topic"
  ) + theme_minimal() + theme(legend.position = "blank")

#DOTPLOT
(p <- ggplot(df_time, aes(x=relative_time, y = PNUM, color=fct_rev(code_topic))) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time$TASK) +
  scale_color_brewer(type="qual", palette = 3) +
  theme_minimal() + labs(
    title = "Topic of Utterances over timecourse of Task",
    x= "timecourse of task (seconds)", y = "Task",
    color = "Topic"
  ))

#DOTPLOT
ggplot(df_time, aes(x=relative_time, y = fct_rev(TASK), color=fct_rev(code_topic))) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time$PNUM) +
  # facet_grid(df_time$TASK ~ df_time$DATASET) +
  scale_color_brewer(type="qual", palette = 3) +
  theme_minimal() + labs(
    title = "Topic of Utterances over timecourse of Task",
    x= "timecourse of task (seconds)", y = "Participant",
    color = "Topic"
)

ggsave(p, file="figures/topics_in_time.png")

## Saving 7 x 5 in image

2.3 [DETAIL of] Utterances

NEXT we explore the distribution of specific detail utterances across Analysis Task, Dataset, Participant and Time.

RQ: What specific things did participants talk aloud during EDA? Are there any details folks only mention during static(v)interactive, or nominal(v)numeric tasks? Any substantial changes in proportion by TASK or DATASET?

Answer:

2.3.1 DETAIL—PROCESS

#PREP DATA FRAMES
df_process <- df_coded %>% 
  filter(code_topic=="PROCESS") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,code_detail)

df_time_process <- df_time %>% 
  filter(code_topic=="PROCESS") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,relative_time,code_detail)

df_summary_task <- df_process %>% 
  group_by(code_detail, TASK) %>% 
  dplyr::summarise(c = n())

df_summary_dataset <- df_process %>% 
  group_by(code_detail, DATASET) %>% 
  dplyr::summarise(c = n())


#DETAILS BY TASK
ggplot(df_summary_task, aes(x = TASK, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = "PuRd") +
  labs( title = "PROCESS Utterances by TASK",
        subtitle = "",
        caption = "weak to moderate difference in PROCESS utterances by TASK, \n but these do not seem substantial when broken into the two categories",
        x= "TASK", y = "count") + theme_minimal()

#DETAILS BY DATASET
ggplot(df_summary_dataset, aes(x = DATASET, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = "PuRd") +
  labs( title = "PROCESS Utterances by DATASET",
        subtitle = "",
        caption = "much more substantial differences in PROCESS utterances by DATASET, \n consistent with intution Ps had more to say about the numeric (vs) nominal outcome variable",
        x= "DATASET", y = "count") + theme_minimal()

#DETAILS DOTPLOT
ggplot(df_time_process, aes(x=relative_time, y = PNUM, color=fct_rev(code_detail))) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time_process$TASK) +
  scale_color_brewer(type="seq", palette = "PuRd") +
  theme_minimal() + labs(
    title = "PROCESS Utterances by timecourse of Task",
    caption = "appear randomly distributed through time \n expected and reasonable given PROCESS utterances are meta-level",
    x= "timecourse of task (seconds)", y = "Participant",
    color = "Topic"
  )

#DETAIL HISTOGRAMS BY TASK
ggplot(df_time_process, aes(x = relative_time, fill = fct_rev(code_detail))) + 
  geom_histogram(binwidth = 30) + 
  facet_grid(df_time_process$TASK ~ df_time_process$code_detail ) +
  scale_fill_brewer(type="seq", palette = "PuRd") +
  theme_minimal() + labs(
    title = "PROCESS Utterances by timecourse of Task",
    x= "timecourse of task (seconds)", y = "frequency of utterances"
  ) + theme_minimal() + theme(legend.position = "blank")

#PROCESSES by PARTICPANT facet TASK
(p <- gf_bar( PNUM ~., fill = ~ fct_rev(code_detail), data = df_process) %>% 
  gf_facet_grid(.~TASK) + 
  scale_fill_brewer(type="seq", palette = "PuRd") +
  labs(
    title = "PROCESS Utterances by Participant and Task",
    subtitle = "",
    caption = "TODO explore P13 representation comments",
    x = "number of coded utterances",
    y = "participant",
    fill = "TOPIC"
  ) + theme_minimal()
)

ggsave(p, file="figures/process_counts.png")

2.3.2 DETAIL—DATASET

#PREP DATA FRAMES
df_dataset <- df_coded %>% 
  filter(code_topic=="DATASET") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,code_detail)

df_time_dataset <- df_time %>% 
  filter(code_topic=="DATASET") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,relative_time,code_detail)

df_summary_task <- df_dataset %>% 
  group_by(code_detail, TASK) %>% 
  dplyr::summarise(c = n())

df_summary_dataset <- df_dataset %>% 
  group_by(code_detail, DATASET) %>% 
  dplyr::summarise(c = n())


#DETAILS BY TASK
ggplot(df_summary_task, aes(x = TASK, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = 4) +
  labs( title = "DATASET Utterances by TASK",
        subtitle = "",
        caption = "notable decrease MISSING DATA utterances in IXN \n unsure what might explain this, explore at individual level",
        x= "TASK", y = "count") + theme_minimal()

#DETAILS BY DATASET
ggplot(df_summary_dataset, aes(x = DATASET, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = 4) +
  labs( title = "DATASET Utterances by DATASET",
        subtitle = "",
        caption = "minor differences by DATASET  reasonable given DATASET \n  represenations are typically tabluar (data dictionary, describe, head/tail)",
        x= "DATASET", y = "count") + theme_minimal()

#DETAILS DOTPLOT
ggplot(df_time_dataset, aes(x=relative_time, y = PNUM, color=fct_rev(code_detail))) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time_dataset$TASK) +
  scale_color_brewer(type="seq", palette = 4) +
  theme_minimal() + labs(
    title = "DATASET Utterances by timecourse of Task",
    x= "timecourse of task (seconds)", y = "Participant",
    caption = "notable sparsity in center of timecourse, reasonable as EDA normative behavior \n is to consider dataframe shape and missing data at the start of an analysis",
    color = "Topic"
  )

#DETAIL HISTOGRAMS BY TASK
ggplot(df_time_dataset, aes(x = relative_time, fill = fct_rev(code_detail))) + 
  geom_histogram(binwidth = 30) + 
  facet_grid(df_time_dataset$TASK ~ df_time_dataset$code_detail ) +
  scale_fill_brewer(type="seq", palette = 4) +
  theme_minimal() + labs(
    title = "DATASET Utterances by timecourse of Task",
    x= "timecourse of task (seconds)", y = "frequency of utterances",
        caption = "sensical that the most uniformly distributed detail code is missing data \n as this can be discovered via graphing",
  ) + theme_minimal() + theme(legend.position = "blank")

#DATASET UTTERANCES by PARTICPANT facet TASK
(p <- gf_bar( PNUM ~., fill = ~ fct_rev(code_detail), data = df_dataset) %>% 
  gf_facet_grid(.~TASK) + 
  scale_fill_brewer(type="seq", palette = 4) +
  labs(
    title = "DATASET Utterances by Participant and Task",
    subtitle = "",
    x = "number of coded utterances",
    y = "participant",
    fill = "TOPIC",
    caption = "P11, P12 contribute to MISSING DATA \n P9 contributes largely to variable metadata \n INVESTIGATE FURTHER",
  ) + theme_minimal()
)

ggsave(p, file="figures/dataset_counts.png")

2.3.3 DETAIL—VARIABLE

#PREP DATA FRAMES
df_variable <- df_coded %>% 
  filter(code_topic=="VARIABLE") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,code_detail)

df_time_variable <- df_time %>% 
  filter(code_topic=="VARIABLE") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,relative_time,code_detail)

df_summary_task <- df_variable %>% 
  group_by(code_detail, TASK) %>% 
  dplyr::summarise(c = n())

df_summary_dataset <- df_variable %>% 
  group_by(code_detail, DATASET) %>% 
  dplyr::summarise(c = n())


#DETAILS BY TASK
ggplot(df_summary_task, aes(x = TASK, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = 5) +
  labs( title = "VARIABLE Utterances by TASK",
        subtitle = "",
        caption = "notably more RANGE obs in STATIC than INTERACTIVE \n TODO consider collapsing distribution-variance ",
        x= "TASK", y = "count") + theme_minimal()

#DETAILS BY DATASET
ggplot(df_summary_dataset, aes(x = DATASET, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = 5) +
  labs( title = "VARIABLE Utterances by DATASET",
        caption = "Notably more SHAPE in SPACE than HAPPINESS \n notably fewer RANGE in SPACE than HAPPINESS \n TODO are shape and range normalized across variable types? ",
        subtitle = "",
        x= "DATASET", y = "count") + theme_minimal()

#DETAILS DOTPLOT
ggplot(df_time_variable, aes(x=relative_time, y = PNUM, color=fct_rev(code_detail))) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time_variable$TASK) +
  scale_color_brewer(type="seq", palette = 5) +
  theme_minimal() + labs(
    title = "VARIABLE Utterances by timecourse of Task",
    caption = "IXN appears more BIMODAL than STATIC where the distribution is more uniform",
    x= "timecourse of task (seconds)", y = "Participant",
    color = "Topic"
  )

#DETAIL HISTOGRAMS BY TASK
ggplot(df_time_variable, aes(x = relative_time, fill = fct_rev(code_detail))) + 
  geom_histogram(binwidth = 30) + 
  facet_grid(df_time_variable$TASK ~ df_time_variable$code_detail ) +
  scale_fill_brewer(type="seq", palette = 5) +
  theme_minimal() + labs(
    title = "VARIABLE Utterances by timecourse of Task",
    x= "timecourse of task (seconds)", y = "frequency of utterances"
  ) + theme_minimal() + theme(legend.position = "blank")

#VARIABLE UTTERANCES by PARTICPANT facet TASK
(p <- gf_bar( PNUM ~., fill = ~ fct_rev(code_detail), data = df_variable) %>% 
  gf_facet_grid(.~TASK) + 
  scale_fill_brewer(type="seq", palette = 5) +
  labs(
    title = "VARIABLE Utterances by Participant and Task",
    subtitle = "",
    x = "number of coded utterances",
    y = "participant",
    fill = "TOPIC",
    caption = "substantial individual differences, most everyone made some comments \n at some point about distribution shape, but \n discussion of outliers, variance and range was more idiosyncratic \n TODO INVESTIGATE P4 STATIC RANGE very high",
  ) + theme_minimal()
)

ggsave(p, file="figures/variable_counts.png")

2.3.4 DETAIL—RELATIONSHIP

#PREP DATA FRAMES
df_relationship <- df_coded %>% 
  filter(code_topic=="RELATIONSHIP") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,code_detail)

df_time_relationship <- df_time %>% 
  filter(code_topic=="RELATIONSHIP") %>% 
  dplyr::select(pid,PNUM,TASK,DATASET,relative_time,code_detail)

df_summary_task <- df_relationship %>% 
  group_by(code_detail, TASK) %>% 
  dplyr::summarise(c = n())

df_summary_relationship <- df_relationship %>% 
  group_by(code_detail, DATASET) %>% 
  dplyr::summarise(c = n())


#DETAILS BY TASK
ggplot(df_summary_task, aes(x = TASK, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = 3) +
  labs( title = "RELATIONSHIP Utterances by TASK",
        subtitle = "",
        caption = "TODO think about proportion of existence and strength/direction",
        x= "TASK", y = "count") + theme_minimal()

#DETAILS BY DATASET
ggplot(df_summary_relationship, aes(x = DATASET, y=c, fill= code_detail)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="seq", palette = 3) +
  labs( title = "RELATIONSHIP Utterances by DATASET",
        subtitle = "",
        caption = "substantial differences by DATASET, \n consistent with pattern of results with VARIABLE utterances \n perhaps due to sparsity of knowledge of analysis of nominal X nominal relationships, and tool coverage",
        x= "DATASET", y = "count") + theme_minimal()

gf_bar( ~code_detail, fill = ~code_detail, data = df_relationship) %>% 
  gf_facet_grid(TASK ~ DATASET) + labs(
    caption = "TODO CONSIDER THIS",
  )

#DETAILS DOTPLOT
ggplot(df_time_relationship, aes(x=relative_time, y = PNUM, color=fct_rev(code_detail))) + 
  geom_point(alpha=0.5, size=3) +
  facet_grid(df_time_relationship$TASK) +
  scale_color_brewer(type="seq", palette = 3) +
  theme_minimal() + labs(
    title = "RELATIONSHIP Utterances by timecourse of Task",
    caption = "uniformly distributed, as expected \n may see variance if dimension of HYPOTHESIS vs OBSERVATION was coded",
    x= "timecourse of task (seconds)", y = "Participant",
    color = "Topic"
  )

#DETAIL HISTOGRAMS BY TASK
ggplot(df_time_relationship, aes(x = relative_time, fill = fct_rev(code_detail))) + 
  geom_histogram(binwidth = 30) + 
  facet_grid(df_time_relationship$TASK ~ df_time_relationship$code_detail ) +
  scale_fill_brewer(type="seq", palette = 3) +
  theme_minimal() + labs(
    title = "RELATIONSHIP Utterances by timecourse of Task",
    x= "timecourse of task (seconds)", y = "frequency of utterances"
  ) + theme_minimal() + theme(legend.position = "blank")

#RELATIONSHIP UTTERANCES by PARTICPANT facet TASK
(p <- gf_bar( PNUM ~., fill = ~ fct_rev(code_detail), data = df_relationship) %>% 
  gf_facet_grid(.~TASK) + 
  scale_fill_brewer(type="seq", palette = 3) +
  labs(
    title = "RELATIONSHIP Utterances by Participant and Task",
    subtitle = "",
    x = "number of coded utterances",
    y = "participant",
    fill = "TOPIC"
  ) + theme_minimal()
)

ggsave(p, file="figures/relationship_counts.png")

3 REPRESENTATIONS

3.1 Representation Groups

How many representations were created?

print("BY TASK")

[1] “BY TASK”

ctable(df_coded$reps_group, df_coded$TASK,
    prop = "t",
    plain.ascii = FALSE)

3.1.1 Cross-Tabulation, Total Proportions

3.1.1.1 reps_group * TASK

Data Frame: df_coded

	TASK	static	ixn	Total
reps_group
barplot		8 ( 1.1%)	8 ( 1.1%)	16 ( 2.2%)
columns		4 ( 0.5%)	0 ( 0.0%)	4 ( 0.5%)
columns_data_dictionary		1 ( 0.1%)	0 ( 0.0%)	1 ( 0.1%)
data_dictionary		27 ( 3.6%)	29 ( 3.9%)	56 ( 7.5%)
data_dictionary_dataframe		1 ( 0.1%)	0 ( 0.0%)	1 ( 0.1%)
data_dictionary_describe		9 ( 1.2%)	0 ( 0.0%)	9 ( 1.2%)
dataframe		50 ( 6.7%)	26 ( 3.5%)	76 ( 10.2%)
dataframe_describe		1 ( 0.1%)	0 ( 0.0%)	1 ( 0.1%)
dataframe_heatmap		0 ( 0.0%)	1 ( 0.1%)	1 ( 0.1%)
dataframe_pairplot		1 ( 0.1%)	0 ( 0.0%)	1 ( 0.1%)
dataframe_profile		1 ( 0.1%)	2 ( 0.3%)	3 ( 0.4%)
describe		11 ( 1.5%)	12 ( 1.6%)	23 ( 3.1%)
describe_profile		1 ( 0.1%)	0 ( 0.0%)	1 ( 0.1%)
double-profiler		16 ( 2.2%)	7 ( 0.9%)	23 ( 3.1%)
heatmap		12 ( 1.6%)	7 ( 0.9%)	19 ( 2.6%)
hist		6 ( 0.8%)	0 ( 0.0%)	6 ( 0.8%)
info		9 ( 1.2%)	4 ( 0.5%)	13 ( 1.8%)
lineplot		36 ( 4.9%)	0 ( 0.0%)	36 ( 4.9%)
Multi-view Chart		8 ( 1.1%)	51 ( 6.9%)	59 ( 8.0%)
none		24 ( 3.2%)	19 ( 2.6%)	43 ( 5.8%)
pairplot		38 ( 5.1%)	12 ( 1.6%)	50 ( 6.7%)
profile		47 ( 6.3%)	59 ( 8.0%)	106 ( 14.3%)
python		37 ( 5.0%)	23 ( 3.1%)	60 ( 8.1%)
scatterplot		49 ( 6.6%)	79 (10.6%)	128 ( 17.3%)
stripplot		6 ( 0.8%)	0 ( 0.0%)	6 ( 0.8%)
Total		403 (54.3%)	339 (45.7%)	742 (100.0%)

print("BY DATASET")

[1] “BY DATASET”

ctable(df_coded$reps_group, df_coded$DATASET,
    prop = "t",
    plain.ascii = FALSE)

3.1.2 Cross-Tabulation, Total Proportions

3.1.2.1 reps_group * DATASET